Virtual Clinical Studies to Examine the Probability Distribution of … · 2017. 8. 25. · for...

RESEARCH PAPER

Virtual Clinical Studies to Examine the Probability Distributionof the AUC at Target Tissues Using Physiologically-BasedPharmacokinetic Modeling: Application to Analyses of the Effectof Genetic Polymorphism of Enzymes and Transporterson Irinotecan Induced Side Effects

Kota Toshimoto1& Atsuko Tomaru1 & Masakiyo Hosokawa2 & Yuichi Sugiyama1

Received: 31 December 2016 /Accepted: 28 March 2017 /Published online: 10 April 2017# The Author(s) 2017. This article is published with open access at SpringerLink.com

ABSTRACTPurpose To establish a physiologically-based pharmacokinet-ic (PBPK) model for analyzing the factors associated with sideeffects of irinotecan by using a computer-based virtual clinicalstudy (VCS) because many controversial associations betweenvarious genetic polymorphisms and side effects of irinotecanhave been reported.Methods To optimize biochemical parameters of irinotecanand its metabolites in the PBPK modeling, a Cluster Newtonmethod was introduced. In the VCS, virtual patients weregenerated considering the inter-individual variability and ge-netic polymorphisms of enzymes and transporters.Results Approximately 30 sets of parameters of the PBPKmodel gave good reproduction of the pharmacokinetics ofirinotecan and its metabolites. Of these, 19 sets gave relativelygood description of the effect of UGT1A1 *28 and SLCO1B1c.521T>C polymorphism on the SN-38 plasma concentra-tion, neutropenia, and diarrhea observed in clinical studiesreported mainly by Teft et al. (Br J Cancer. 112(5):857-65,20). VCS also indicated that the frequency of significant asso-ciation of biliary index with diarrhea was higher than that ofUGT1A1 *28 polymorphism.Conclusion The VCS confirmed the importance of geneticpolymorphisms of UGT1A1 *28 and SLCO1B1 c.521T>C

in the irinotecan induced side effects. The VCS also indicatedthat biliary index is a better biomarker of diarrhea thanUGT1A1 *28 polymorphism.

KEY WORDS Cluster Newtonmethod . inter-individualvariability . irinotecan(CPT-11) . neutropenia .physiologically-based pharmacokinetic modeling .virtual clinical study

ABBREVIATIONSAUC Area under the curve of concentration time profileBCRP Breast cancer resistance proteinCES CarboxylesteraseCL ClearanceCNM Cluster Newton methodCYP Cytochrome P450IgG Immunoglobulin GMDR1 Multiple drug resistance 1MRP2 Multidrug resistance-associated protein 2OATP Organic anion transporting polypeptidePBPK Physiologically-based pharmacokineticsPS Permeability surface area-productSN-38 7-Ethyl-10-hydroxy-camptothecinSN-38G 7-Ethyl-10-hydroxy-camptothecin glucuronideUGT Uridine diphosphate glucuronosyltransferasesVCS Virtual clinical studyWSS Weighted sum of squares

INTRODUCTION

Irinotecan (CPT-11) is a prodrug for the treatment of variouskinds of cancer. Irinotecan exerts its pharmacological effect byblocking DNA replication and transcription by topoisomerase-

Electronic supplementary material The online version of this article(doi:10.1007/s11095-017-2153-z) contains supplementary material, which isavailable to authorized users.

* Yuichi [email protected]

1 Sugiyama Laboratory, RIKEN Innovation Center, RIKEN1-7-22 Suehiro-cho, Tsurumi-ku, Yokohama, Kanagawa 230-0045, Japan

2 Faculty of Pharmaceutical Sciences, Chiba Institute of ScienceChoshi, Japan

Pharm Res (2017) 34:1584–1600DOI 10.1007/s11095-017-2153-z

http://orcid.org/0000-0002-8452-8835

http://dx.doi.org/10.1007/s11095-017-2153-z

http://crossmark.crossref.org/dialog/?doi=10.1007/s11095-017-2153-z&domain=pdf

1 inhibition (1). Irinotecan has a complex metabolic pathwayin which carboxylesterase (CES) 1 and CES2 in the liver, gas-trointestinal tract, and plasma convert irinotecan to its activemetabolite SN-38 as a first step. SN-38 is then metabolized tothe inactive metabolite SN-38 glucuronide (SN-38G) by uri-dine diphosphate glucuronosyltransferases (UGTs), especiallyUGT1A1. Deconjugation from SN-38G to SN-38 occurs byintestinal β-deglucuronidase. In addition, irinotecan undergoesCYP3A4/5 metabolism to inactive metabolites APC, M4, andNPC with further conversion to SN-38 by CES1 and CES2(Fig. 1) (2).

The severe adverse effects of irinotecan are neutropeniaand delayed diarrhea, which are caused by high exposureof plasma and intestinal epithelia to SN-38 (3). Irinotecanhas been widely studied by clinical pharmacogenomics fo-cused on UGT1A1 (4,5). A UGT1A1 *28 or UGT1A1 *6phenotype may increase the risk of neutropenia owing tohigher SN-38 exposure by decreasing the UGT1A1 meta-bolic clearance, as found in various clinical studies (6,7). Inaddition to the UGT1A1, some transporter genotypes arereported as having key roles in the adverse effects ofIrinotecan. Decreased uptake clearance among patientswith an SLCO1B1 (OATP1B1) c.521T>C genotypecauses severe irinotecan toxicity (8). Although other clinicalstudies report that ABC transporter genotypes such asABCB1 (MDR1), ABCG2 (BCRP), and ABCC2 (MRP2),which are involved in biliary excretion and intestinal effluxin the intestinal lumen, have been linked to exposure toSN-38 and the side effects of irinotecan (9,10), others havedescribed that these genetic polymorphisms have no asso-ciation with the adverse effects (11,12). Thus, differentfindings by various clinical studies have been made withregard to the associations between enzymes/transportersand drug-induced side effects (neutropenia, diarrhea).This is probably because insufficient numbers of patientswere included in most of those previous clinical studies.

By contrast, to predict drug pharmacokinetics quantita-tively, physiologically-based pharmacokinetic (PBPK)modeling has been more recently used to predict bloodand tissue concentration–time profiles. Over 350 articleson human PBPK modeling have been published since2008 (13). The number of regulatory submissions to theU.S. FDA using PBPK model analysis has gradually in-creased since 2008 (14). Furthermore, the potential of vir-tual clinical study (VCS), which uses computational simu-lation with PBPK, pharmacodynamic, and toxicodynamicmodels—with virtual patients generated by the variabilityof physiological and pharmacokinetic parameters based ongenetic polymorphism, ethnic differences, and inter- andintra-individual variability—is an advantage of PBPKmodel analysis (15). The concept of VCS has already beenintegrated into PBPK model analysis, where the associa-tion between SLCO1B1 genotype and rosuvastatin phar-macological effect (a decline of the plasma concentration ofmevalonic acid) has been reported (16). In VCS, since thevalues of all parameters for each virtual patient are known,it is possible to identify the factors significantly associatedwith the effects and side effects based on the large numberof simulation results. It was considered that an accuratePBPK model of irinotecan and its metabolites would en-able us to identify the factors related to its side effects. Abrief PBPK model of irinotecan and SN-38 metabolismhas been published by Fujita et al. (17). They predictedSN-38 exposure after renal failure. However, because sucha simple PBPK model did not contain an enterocyte com-partment, it was impossible to predict diarrhea.

The purpose of the present study was to gain insight intothe factors highly associated with neutropenia and diarrheausing a VCS approach with the PBPK model. The parame-ters for the PBPK model, which could reproduce the averageblood concentration–time profile of irinotecan and its metab-olites, were estimated as the first step of our analyses. Becausethere were 46 unknown parameters within the search space, itwas difficult to set appropriate initial values for unknown pa-rameters; thus, conventional nonlinear least squares methodssuch as the Gauss–Newton method and Levenberg–Marquardt method could not be applied. To overcome thisproblem, a Cluster Newton method (CNM) was newlyemployed to optimize unknown parameters (18). In this algo-rithm, multiple sets of initial values of parameters were firstgenerated at random from a certain range of initial values foreach unknown parameter. Then, linear approximations ofprojection from parameters to objective function were usedto determine the next iterations from initial sets. CNM hassome advantages over conventional parameter optimizationalgorithms (19). The first advantage is that it is easy to set theinitial condition of the parameters. While CNM only requiresa rough range of values for each parameter, many conven-tional approaches require fixed initial possible values. TheFig. 1 Metabolic and transport pathway for irinotecan (CPT-11).

Virtual Clinical Study Analyses of Irinotecan Induced Side Effect 1585

second advantage is that multiple optimized sets of parameterscan be obtained using a single optimization.

In the present study, 46 unknown parameters in the PBPKmodel of the metabolism of irinotecan and its metaboliteswere optimized by CNM. Then, obtained sets of parameterswere evaluated regarding whether the PBPK model with op-timized parameters could reproduce the outcomes of clinicalstudies, focusing on the effect of UGT1A1 *28 and SLCO1B1c.521T>C polymorphism, as reported by Teft et al. (20) and insome other clinical studies, and the reproduction of clinicaloutcomes was attempted using a VCS approach.

MATERIALS AND METHODS

Measurement of the Contribution of CES1 Metabolismof Irinotecan to SN-38

The possible involvement of CES1 in the metabolism ofirinotecan in the liver was assessed using anti-CES1A IgG.The methods were as follows.

CES1A IgG-treated microsomes were prepared by a mod-ification of a procedure described previously (21). Microsomes(50 mixed-sex donors, Corning, Japan) were solubilized with0.5% cholic acid in 10 mM Tris-HCl buffer (pH 8.0). Aftersolubilizing, this mixture was centrifuged, and anti-CES1AIgG was added to the supernatant and incubated for 30 minat 37°C, and then for 24 h at 4°C. After the incubation,nProtein A Sepharose 4 Fast Flow (GE Healthcare, Japan)was added to the mixture, which was then incubated for 1 hon ice. Then, this mixture was centrifuged and the hydrolaseactivity in the supernatant was determined.

The method for assessment of hydrolase activity was mod-ified from a procedure described previously (22). To assesshydrolase activity, temocapril was used as a positive controlfor CES1 activity. Temocapril (1 μM) and irinotecan (2 μM)were added to the CES1A IgG-treated or control microsomes(0.1 mg/mL) and incubated at 37°C for up to 30 min. Afterthe incubation, the reaction was stopped by the addition of0.1% formic acid in acetonitrile containing an internal stan-dard. The generation of metabolites (SN-38 for irinotecanand temocaprilat for temocapril) was measured by LC-MS/MS. The LC-MS/MS consisted of a Nexera X2 separatingmodule (Shimadzu, Kyoto, Japan) and an LCMS8050 massspectrometer (Shimadzu) with an electron ion spray interface.The mass spectrometer was operated in a multiple reactionmonitoring mode using the respective MH+ ions, m/z 587 tom/z 124 for irinotecan, m/z 393 to m/z 349 for SN-38, m/z477 to m/z 270 for temocapril, m/z 449 to m/z 270 fortemocaprilat, and m/z 277 to m/z 175 for chlorpropamide(internal standard). Chromatographic separation was per-formed on a C18 column (Kintex C18, 2.1 × 100 mm,2.6 μm; Phenomenex Inc., Torrance, CA) at 40°C under

gradient conditions (0.1% formic acid and acetonitrile) at0.2 mL/min. The stabilities of SN-38 and temocaprilat wereconfirmed up to 20 h, and no degradation was observed dur-ing the procedure.

Development of the PBPK Model of Irinotecan and ItsMetabolites

The PBPK model of irinotecan and its metabolites SN-38,SN-38G, NPC, and APC is shown in Fig. 2. A module ofthe PBPK model shown in Fig. 2a was developed and con-nected as depicted in Fig. 2b. For each module, the livercompartment was divided into five units of extrahepatic andhepatocyte compartments to mimic a dispersionmodel (5 livermodel (23)) and take the permeability-limited process of he-patic uptake into account. As biliary excretion of irinotecanand its metabolites was observed from a mass balance study ofa patient with a biliary T-tube (24), three transit compart-ments between hepatocytes and the intestinal lumen wereincorporated to represent enterohepatic circulation. Thestructure of a segregated flow model (25) that could representintestinal metabolism and transport was employed.

Fixed physiological and physicochemical parameterswere obtained from the literature and our experiments(Supplementary Table 1). Tissue-to-blood concentrationratios (Kp,muscle, Kp,skin, Kp,adipose, and Kp,gut), and proteinunbound fraction in the liver (fh) and enterocytes (fgut) werebased on reported in silico calculations using clogP andpKa (acidic and basic) obtained from SciFinder Scholar(Chemical Abstracts Service, Columbus, OH) (26,27).The name and initial range of unknown parameters forCNM optimization are described in Table I. In the livercompartment, parameter optimization with some hybridparameters (CLint,all, Rdif,h, β, fbile) (28) was performed,except for irinotecan. PSact,inf,h was defined as the hepaticactive uptake via transporters, PSdif,inf,h as the influx clear-ance by passive diffusion through sinusoidal membrane inthe liver, PSdif,eff,h as the efflux clearance by passive diffu-sion through the sinusoidal membrane in the liver, CLmet,h

as the metabolic clearance in the liver, and CLbile as thebiliary excretion clearance. Then, the overall intrinsicclearance (CLint,all) was described in Eq. 1 (29).

CLint;all ¼ PSact;in f ;h þ PSdi f ;in f ;h� �

⋅CLmet;h þ CLbile

PSdi f ;e f f ;h þ CLmet;h þ CLint;bileð1Þ

In the present research, PSdif,inf,h = PSdif,eff,h was assumed.Rdif,h, β, and fbile were defined as described in Eqs. 2, 3, and 4.

Rdi f ;h ¼ PSdi f ;in f ;hPSact;in f ;h

ð2Þ

1586 Toshimoto, Tomaru, Hosokawa and Sugiyama

β ¼ CLmet;h þ CLint;bile

PSdi f ;e f f ;h þ CLmet;h þ CLint;bileð3Þ

f bile ¼CLbile

CLmet;h þ CLbileð4Þ

As shown in Fig. 1, SN-38 is considered to be taken up byOATP1B1 into the liver based on a previous report (30). Theuptake study of SN-38G was performed using anHEK293/OATP1B1-expressing cell system. The amount ofSN-38G in the cell was measured by LC-MS/MS which wasused the measurement of the contribution of CES1 metabo-lism of irinotecan to SN-38. The uptake of SN-38G byHEK293/OATP1B1 cells was showed time-dependent in-crease and higher than that by HEK293/mock cells(Supplementary Fig. 1). Therefore, SN-38G is also consideredwith OATP1B1-mediated uptake in the liver. No involvementof active uptake transporters for other compounds wasassumed.

In the intest inal compartment, the value ofPSdif,eff,ent, which represents efflux clearance by passivediffusion through a basolateral membrane from theenterocyte to the mucosal blood, and PSdif,inf,ent, whichrepresents influx clearance by passive diffusion throughbasolateral membranes from the mucosal blood to theenterocyte, was assumed to be equal. The value of pas-sive diffusional clearance from the enterocyte to the in-testinal lumen was defined as the product of PSdif,eff,entand an apical/basolateral area ratio (AR = 20 (31)).

PSact,eff,ent represents the active efflux transporters viaABCB1, ABCG2, and ABCC2 in the intestine. A hy-brid parameter Rdif,ent was defined as the ratio of pas-sive diffusional efflux to the active efflux in theenterocytes as described in Eq. 5 for compounds thatare substrates of these efflux transporters.

Rdif ;ent ¼ AR* PSdif ;eff ;entPSact;eff ;ent

ð5Þ

The differential equations of the PBPK model are de-scribed in the Supplementary Text.

The process employed to optimize parameters using CNMwas as follows.

1. 10,000 initial sets of parameters, the range of which isdescribed in Table I for each unknown parameter, weregenerated.

2. CNM was used to optimize the 10,000 sets of pa-rameters. To provide the diversity of the problemsolutions (optimized sets of parameters), AUC foreach compound was used as the objective functionfor optimization. Only stage 1 of the CNM optimi-zation (18) was performed to avoid overfitting andsave computational time.

3. Procedures 1 and 2 described above were repeated 100times with 10,000 different initial sets of parameters.

4. The blood concentration–time profile of irinotecanand its metabolites was simulated using 1,000,000optimized sets of parameters. The weighted sum ofsquares (WSS) between the simulated and observed

Fig. 2 Structures of the physiologically-based pharmacokinetic (PBPK) model of irinotecan and its metabolites. (a) shows a module of PBPK model. The livercompartment was adapted as a permeability-limited model and divided into five compartments (23) to construct a model similar to the dispersion model.Enterohepatic circulation was considered. Permeability between the intestinal lumen and enterocytes and between enterocytes and mucosal blood was alsoincorporated. Portal vein blood flow was divided into mucosal blood flow (Qmucosa) and serosal blood flow (Qserosa). (b) Each module for irinotecan and themetabolites is connected shown here based on the scheme in Fig. 1.


blood concentration–time profile described in Eq. 6was calculated

WSS ¼ ∑ni¼1

yi−y′i� �2

y2i; yi : ith observed value; y′i : ith predicted value ð6Þ

5. The top 30 sets of parameters that minimized the WSSbetween simulated and observed concentration–time pro-files at the sampling points were selected.

The clinical study reported by van der Bol et al. (32)was used for observed AUC and blood concentration–

time profiles of irinotecan, SN-38, SN-38G, NPC, andAPC. 30 selected sets of parameters were used for VCSas described below and given in ascending order ofWSS for IDs 1 to 30.

Generation of Virtual Patients

To perform VCS, consideration of genetic polymorphisms ofsome enzymes and transporters is required, in addition to theinter-individual variability for each parameter of the PBPKmodel. This VCS considered 6 genetic polymorphisms of

Table I Initial Range and Theoretical Lower Limit of Parameters for CNM

A. Common parameters

Parameter Unit Irinotecan SN-38 SN-38G NPC APC Theoreticallower limit

Vcentral L/kg 0.075 ~ 0.75 0.075 ~ 0.75 0.075 ~ 0.75 0.075 ~ 0.75 0.075 ~ 0.75 0.074a

ka /h 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0

kfeces /h 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0

Rdif,h - -b,c 0.01 ~ 1.0 0.01 ~ 1.0 -c -c 0

1/β - -b 1.1 ~ 10 1.1 ~ 10 1.1 ~ 10 1.1 ~ 10 1

CLint,all L/h/kg -b 0.1 ~ 100 0.1 ~ 100 0.1 ~ 100 0.1 ~ 100 0

1/fbile - -b -e -d 1.1 ~ 10 -d 1

kbile /h 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0.06 ~ 6.0 0

Rdif,ent - 0.1 ~ 10 0.1 ~ 10 0.1 ~ 10 -c -c 0

B. Irinotecan parameters

Parameter Unit Value Theoretical lower limit

PSdif,inf,h L/kg 0.1 ~ 100 0

CLSN-38,h(irinotecan) L/h/kg 0.1 ~ 100 0

CLNPC,h(irinotecan) L/h/kg 0.1 ~ 100 0

CLAPC,h(irinotecan) L/h/kg 0.1 ~ 100 0

CLothers,h(irinotecan) L/h/kg 0.1 ~ 100 0

CLbile L/h/kg 0.1 ~ 100 0

PSdif,eff,ent L/h/kg 0.1 ~ 100 0

CLSN-38,h(irinotecan) /CLSN-38,ent(irinotecan) - 1.1 ~ 10 1

CLNPC,h(irinotecan) / CLNPC,ent(irinotecan) - 1.1 ~ 10 1

C. SN-38 parameters


1/fglue - 1.1 ~ 10 1

CLSN-38G,h(SN-38) /CLSN-38G,ent(SN-38) - 1.1 ~ 10 1

D. SN-38G parameters


kdec /h 0.06 ~ 6 0

a Vcentral is larger than blood volumebHybrid parameters were not used. Instead of it, as shown in Table 1B. parameters for representing elementary steps were usedcNo involvement of active uptake or efflux transporters for these compoundsd fbile value was fixed as 1e The ratio (fglu) of metabolic clearance for glucuronidation of SN-38 to hepatic intrinsic clearance was estimated using the following equation, fglu = 1 - fblie


enzyme and transporters (UGT1A1*28, SLCO1B1 c.521T>Cand c.388A>G, ABCG2 c.421C>A, ABCB1 c.3435C>T, andABCC2 c.-24C>T). The contribution of each enzyme andtransporter in the metabolism and transporter process is as-sumed as follows: UGT1A1, metabolism from SN-38 to SN-38G in the liver (CLSN-38G,h(SN-38)) and intestine (CLSN-

38G,ent(SN-38)), contributions for both are 100%; SLCO1B1, he-patic active uptake of SN-38 and SN-38G (PSact,inf,h(SN-38) andPSact,inf,h(SN-38G)), contributions for both are 100%;ABCG2, biliary excretion of SN-38 (CLbile(SN-38)) andactive efflux from enterocytes to the intestinal lumenof SN-38 (PSact,eff,ent(SN-38)), contributions for both are33%; ABCB1, biliary excretion of irinotecan and SN-38 (CLbile(irinotecan) and CLbile(SN-38)) and active efflux ofirinotecan and SN-38 from enterocytes to the intestinallumen (PSact,eff,ent(irinotecan) and PSact,eff,ent(SN-38)), contri-butions are 50% for irinotecan and 33% for SN-38; andABCC2, biliary excretion of irinotecan, SN-38, and SN-38G(CLbile(irinotecan), CLbile(SN-38), and CLbile(SN-38G)) and active ef-flux of irinotecan, SN-38, and SN-38G from enterocytes tothe intestinal lumen (PSact,eff,ent(irinotecan), PSact,eff,ent(SN-38), andPSact,eff,ent(SN-38G)), contributions are 50% for irinotecan, 33%for SN-38, and 100% for SN-38G. SupplementaryTable 2 shows the activity ratio and the allele frequencyfor each genetic polymorphism obtained from experi-ments in vitro and in vivo.

A virtual patient was generated by performing Monte Carlosimulation following a given frequency distribution (average,coefficient of variation [CV], and the shape of distribution) foreach parameter. Hybrid parameters (e.g. CLint,all, β, Rdif,h, fbile,Rdif,ent) used for CNM optimization were converted to eachparameter representing elementary processes. The averages ofphysiological parameters are presented in SupplementaryTable 3A. Other parameters were obtained by optimizationusing the CNM, and the average values were used together withtheir variabilities. The variabilities (CV) and the shape of distri-bution are described in Supplementary Table 3B. Because theobserved blood concentration–time profile of irinotecanand its metabolites (32) used by the CNM was obtainedfrom patients recruiting the same number of UGT1A1*1/*1 and *1/*28 patients, the value of metabolic clear-ance of SN-38 to SN-38G (CLSN-38G,h(SN-38) and CLSN-

38G,ent(SN-38)) was recalculated for UGT1A1 *1/*1 usingvalues obtained from the CNM as an average valuebetween UGT1A1 *1/*1 and *1/*28.

The variability of each variable in performing MonteCarlo simulation was set as follows.

(1) Body weight and body surface area

The body weight of a virtual patient was generated based onnormal distribution (average and CV are given inSupplementary Table 3). Body surface area (BSA) was

obtained from a normal distribution of which theaverage and CV were calculated from Eq. 7 (33) and5.5%, respectively.

BSA ¼ 4:688⋅BW 0:8168−0:0154⋅log BWð Þð Þ; BSA cm2� �;BW g½ � ð7Þ

CV of 5.5% was determined as follows. After temporaryBSA was calculated from Eq. 7 with the distribution of bodyweight for a healthy man (average, 78.8 kg; CV, 11.7% (34)),actual BSA was calculated for the normal distribution of vari-ability based on a certain set CV. Calculated BSA distributionwas compared with the reported distribution (average, 1.97 m2;CV, 9.6% (35)). CVs estimated using the calculated and report-ed BSA distribution corresponded with each other. In this VCS,the average and CV of body weight distribution were also esti-mated to compare the calculated distribution of BSA with thedistribution reported after clinical study of irinotecan (5,32).

(2) Plasma and blood unbound fraction of compounds

Plasma unbound fraction of compounds was calculatedfrom Eq. 8, which was derived from the Langmuir equation.

f p ¼1

1þ n Pt½ �K d

ð8Þ

In Eq. 8, n represents the number of binding sites of theprotein, [Pt] the protein concentration, and Kd the dissocia-tion constant. fp for each virtual patient was calculated so thatn[Pt]/Kd was generated from the distribution described inSupplementary Table 3. Blood unbound fraction was calcu-lated fromEq. 9 assuming the same unbound concentration ofthe compound between the plasma and blood cells:

f b ¼f p

1− 1−f pf r

r

� �Hct

ð9Þ

where fr represents the unbound fraction in the blood cells andHct the hematocrit. Eq. 8 can be used to calculate fr. In thesecalculations, it was assumed that the inter-individual variabil-ity of n[Pt]/Kd was caused only by the [Pt]. The same [Pt] wasused for different compounds if they were bound to the sameprotein. In the present study, the major protein bound tocompounds in plasma was α1-acid glycoprotein when a com-pound’s basic pKa was over 7 (irinotecan, NPC, and APC),and albumin for the others (SN-38 and SN-38G). In bloodcells, all compounds except for SN-38G are bound to redblood cells. It was assumed that SN-38G was not transferredto blood cells considering Rb reached the theoretical lowerlimit value (1 –Hct). Eq. 10 is used to calculate fb of SN-38G.

f b ¼f pRb

¼ f p1−Hct

ð10Þ


(3) Kp and tissue unbound fraction of compounds

Kp for muscle, skin, adipose, and gut, and tissue unboundfraction (fh and fgut) were obtained using in silico calculations(26,27). pKa, clogP, fp, and fb were used in these calculations.

(4) Renal clearance

Assuming that reabsorption of compounds was negligible,renal clearance (CLr) was calculated using Eq. 11:

CLr ¼ f b⋅GFRþ CLsec ð11Þwhere GFR represents glomerular filtration rate and CLsecsecretion clearance. CLsec is described in Eqs. 12 to 16 basedon the dispersion model.

CLsec ¼ Q rtb 1−4a

aþ 1ð Þ2⋅ea−12DN

� − a−1ð Þ2⋅e

−aþ12DN

� 0BB@

1CCA ð12Þ

a ¼ 1þ 4DN ⋅ f b⋅CLint;sec

Q rtb

� �12

ð13Þ

Q rtb ¼ Q r−GFR ð14Þ

Q r ¼ERPF1−Hct

ð15Þ

ERPF ¼ GFRFF

ð16Þ

where CLint,sec represents intrinsic secretion clearance, Qrtb

renal uriniferous tubule blood flow, Qr renal blood flow,ERPF effective renal plasma flow, and FF filtration fraction.DN is the dispersion number and is set to 0.17. The averageand CV of FF, Hct, and GFR are described in SupplementaryTable 3. The CV of CLint,sec that could reproduce the inter-individual variability of CLr for irinotecan, SN-38, SN-38G,and APC observed in a clinical study (36) was estimated, andtheir average (CV 34.2%), excluding SN-38G, was used forVCS. For SN-38G, the contribution of CLint,sec to CLr is smalland its CV values cannot be calculated.

Performing VCSs and the Comparison of theirOutcomes with Clinical Studies

Using a VCS approach, 30 sets of parameters optimized byCNM so that the blood concentration–time data of irinotecanand its metabolites were well described (33), were evaluatedwhether some sets of parameters can describe well other clin-ical findings [20 and References 1–17 in SupplementaryText]. The clinical findings showing the similarities to the

prediction by VCSs adopted in the present analyses consistof the following 4 categories: (I) the association between SN-38 plasma concentration at the end of infusion (90 min) andUGT1A1 *28 polymorphism or SLCO1B1 c.521T>C poly-morphism; (II) the association between neutropenia and eachgenetic polymorphism of enzymes and transporters; (III) theassociation between diarrhea and each genetic polymorphismof enzymes and transporters; and (IV) the association betweendiarrhea and Bbiliary index (= AUC(irinotecan) × AUC(SN-38) /AUC(SN-38G)),^ which has been reported as a good biomarkerof diarrhea (37). For the categories I, VCS outcomes werecompared with those reported by Teft et al. (20), which iscalled the Btarget study^ in this article. The target study wasperformed with 127 advanced or metastatic cancer patientsand focused on 14 genetic polymorphisms of enzymes andtransporters, including 6 genetic polymorphisms consideredin VCS. For the other categories (Categories II, III, and IV),VCS outcomes were compared not only with the target study,but also with other previously reported studies. The VCSprocedure can be described as follows:

1. A virtual population (n= 127, equivalent to the number oftarget study patients) was generated using processes 1–1and 1–2.

1-1. 6 genetic polymorphisms (UGT1A1 *28, SLCO1B1c.521T>C and c.388A>G, ABCG2 c.421C>A,ABCB1 c.3435C>T, and ABCC2 c.-24C>T) among127 virtual patients were selected.

1-2. A Monte Carlo simulation was used to generate a valuefor each parameter following Supplementary Tables 2and 3.

2. A blood and tissue concentration–time profile was simu-lated, and the probability of side effects was determinedand compared with the findings of the target study.Neutropenia and diarrhea are considered to be relatedto the AUCs in the plasma and intestinal epithelia,respectively.

3. Processes 1 and 2 were repeated 100 times for differentvirtual populations to consider the differences among dif-ferent clinical study results caused by the inter-individualvariability of patients.

4. Processes 1 to 3 were repeated for all 30 sets of parameters.

A Welch’s t test with post hoc Bonferroni correction wasused to compare SN-38 concentration among the polymor-phisms. Neutropenia occurred in 21 and diarrhea in 8 patientsin the target study. To determine the corresponding values ofAUCs in those who suffered from these side effects, we as-sumed that among the 127 virtual patients, those who hadthe highest 21 unbound plasma AUCs for SN-38 developedneutropenia, and those who had the highest 8 unboundenterocyte AUCs for SN-38 developed diarrhea. Althoughside effect factors in the target study were estimated based


on logistic regression, none of the factors used by logistic re-gression were shown in the target study article. Subsequently,Fisher’s extract test was applied to determine the frequency ofpolymorphism between those with and without side effects.Dominant and recessive genetic models were applied inFisher’s extract test. The dominant model compares a combi-nation of heterozygous and homozygous variants to the wildtype. The recessive model compares a homozygous variant toa combination of heterozygous variants and the wild type.The Wilcoxon rank sum test was used was used to determinethe association between the biliary index and diarrhea.

Analysis of the Effect of the Number of Virtual Patientson the Inter-Individual Variability of Clinical Outcomes

The inter-individual variability of patients may affect the resultsof clinical studies with insufficient numbers of patients. Thus,the effect of inter-individual variability among virtual patientsfor the VCS outcomes was compared by performing VCS 100times with different numbers of virtual patients (25, 39, 51, 64,96, 127, 192, 256, 384, 512, 1024, and 1280) (Fig. 6). For eachVCS, the frequency of a significant (p< 0.05) effect of polymor-phism of UGT1A1 *28 or SLCO1B1 c.521T>C on the SN-38plasma concentration at the end of infusion (90 min) was calcu-lated. This analysis was only performed using the set of param-eters IDs 1 and 5 as representatives.

Computational Environment

The simultaneous ordinary differential equations were solvedusing the MATLAB (MathWorks, Natick, MA) stiff ODEsolver ODE15s on MATLAB version 8.6.0 (R2015b).Virtual patients were generated and all statistical tests wereperformed using R version 3.2.3. All calculations were per-formed on a workstation (CPU: Xeon E5–2640 v3 × 2; OS:CentOS 6.7 64 bit; RAM: 32 GB).

RESULTS

Estimation of the Contribution of CES1to the Metabolism of Irinotecan to SN-38

Figure 3 shows the ratio of the production of metabolites(SN-38 from irinotecan and temocaprilat from temocapril)between CES1A IgG-treated and control microsomes (% ofcontrol). Temocapril is a well-known substrate of CES1(38). The ratio for temocapril decreased as the amount ofCES1A IgG increased. By using 0.30 mg CES1A IgG, thevalues were 45.1 ± 2.6 (%) for temocapril and 95.3 ± 2.9(%) for irinotecan compared with controls, indicating thatCES2 plays an important role in the production of SN-38in human liver.

Analyses of Pharmacokinetics of Irinotecan and ItsMetabolites Based on PBPK Modelingand Optimization of Various Biochemical ParametersUsing CNM

Figure 4 shows the blood concentration–time profiles foririnotecan, SN-38, SN-38G, NPC, and APC with 30 sets ofparameters optimized by CNM, which minimizes the WSSbetween simulated and observed concentrations at the varioussampling points. The average WSS for each set of parametersis shown in Supplementary Fig. 2, and optimized values foreach parameter are shown in Supplementary Table 4. Theaverage WSS for the top 30 sets of parameters ranged from0.082–0.107.

Simulation of the Effect of Inter-Individual Variabilityby Using VCSs

The VCS was performed 100 times with a different virtualpopulation for each set of parameters. Figure 5 shows theeffect of inter-individual variability among the virtual popula-tion which comes from other variability than genetic polymor-phism. Figure 5a and b show the effect of UGT1A1 *28 poly-morphism on the plasma concentration of SN-38 at the end ofinfusion (90 min) with different populations (ID 1 and ID 5)and Fig. 5 c and d show the effect of SLCO1B1 c.521T>Cpolymorphism on the plasma concentration of SN-38 at theend of infusion (90 min) with different populations (ID 1 andID 5). In Fig. 5, the top-left panel in each figure shows theresults of the target study (20). The others show 9 of the 100times of VCSs using the set of parameter ID1 (Fig. 5a and c)and ID 5 (Fig. 5b and D). As shown in Fig. 5a, all cases of VCScould reproduce the result of target study using the set of ID 1.However, using the set of ID 5, Fig. 5b shows that only 5 cases

Fig. 3 The contribution of CES1 tometabolism of irinotecan and temocapril.The amount of metabolite normalized by control condition is shown withincreased amount of CES1A-specific IgG (N = 3).


of VCS could reproduce the clinical results. This finding sug-gests that the same clinical outcomes may not be obtainedeven if the clinical studies are repeated when the number ofpatients in the population (n) is small, as in the present case(n = 127) with the set of parameters ID 5. In case of Fig. 5c (ID1) and 5D (ID 5), 3 cases and 6 cases of VCS could reproducethe results of target study for ID 1 and ID 5, respectively.These results suggested that with different set of parameters,different clinical outcomes were predicted.

The Relationship between the Number of VirtualPatients and the Reproduction Frequency of the Effectof Polymorphisms of UGT1A1 and SLCO 1B1on the Plasma Concentration of SN 38.

Figure 6 shows the reproduction frequency of the effect ofUGT1A1 *28 (Fig. 6a) and SLCO1B1 c.521T>C (Fig. 6b)on the plasma SN-38 concentration at the end of infusion(90 min) among 100 times of VCSs for each number of virtualpatient using the set of parameters IDs 1 and 5. The repro-duction frequency increased when the number of virtual pa-tients (n) increased, and was higher than 0.8 for n= 96 with ID1 and n = 192 with ID 5 for UGT1A1 *28 (Fig. 6a) andn = 1024 with ID 1 and n = 192 with ID 5 for SLCO1B1c.521T>C (Fig. 6b).

Comparison of VCS Outcomes among Setsof Parameters Obtained by CNM to the Target Studywith Regard to the Effect of UGT1A1 *28 and SLCO1B1c.521T>C Genotype on Plasma SN-38 Concentration

Figure 7 shows that SN-38 plasma concentration at the end ofinfusion in UGT1A1 *28 (Fig. 7a) and SLCO1B1 c.521T>C(Fig. 7b) genotypes for the target study and 9 VCSs with dif-ferent sets of parameter values optimized by CNM. The topleft shows the clinical results of the target study. The othersshow 9 VCS outcomes in which a different sets of parameterIDs from 1 to 9 were used (Supplementary Table 4). Figure 7aand b show the results of the same sets of parameters for VCSs.In these VCSs, only the outcomes for the set of parameters ID1, 7, and 9 could reproduce the results of the target study withregard to the effects of polymorphism of both UGT1A1 andSLCO1B1 on the SN-38 plasma concentrations.

Selection of the Best Set of Parameters that CouldReproduce the Target Study and Some Other ClinicalStudies with Regard to the Effect of UGT1A1 *28and SLCO1B1 c.521T>C

3000 VCSs (100 VCSs multiplied by 30 sets of parame-ters) were performed. Figure 7 shows only 9 sets of

Fig. 4 Comparison of clinicallyobtained blood concentration–timeprofile of irinotecan (a), SN-38 (b),SN-38G (c), NPC (d), and APC (e)and the simulated lines obtainedusing the top 30 sets of parametersoptimized by CNM. Orange dotsrepresent the observed bloodconcentration (31). Each blue linerepresents the bloodconcentration–time profile using acertain set of optimized parametersobtained using CNM. Each figureshows 30 blue lines that wasobtained by minimizing theweighted sum of squares betweenpredicted and observedconcentrations at all sampling points.


parameters (selected at random) among 30 sets of param-eters. Supplementary Table 5 shows all the results ofVCSs. Outcomes of frequency of statistical significanceof genotypes or phenotype are shown in this table. Somesignificant associations between the genetic polymorphismof enzymes and transporters and SN-38 plasma concen-t ra t i on and s ide e f f e c t s we re p i cked up f romSupplementary Table 5 and was shown in Fig. 8. Theassociation in Fig. 8 satisfied either of the 3 conditionsdescribed below: (I) the association was statistically signif-icant as shown in the target study; (II) the associationbetween the polymorphism of UGT1A1 *28 orSLCO1B1 c.521T>C and side effects (neutropenia, diar-rhea) was statistically significant in other reported studies,

though not in the target study; and (III) the side effectdiarrhea shows an association with biliary index (37).Some VCSs shown in green color indicated the high sim-ilarities between the VCSs and real clinical outcomes. Toexamine whether the 100 times of VCSs using each set ofparameters reproduce the outcomes of clinical studies, thefollowing 7 criteria were used: (I) the effect of UGT1A1*28 heterozygous on plasma concentration of SN-38 wasreproduced over 75 times; (II) the effect of UGT1A1 *28homozygous on plasma concentration of SN-38 wasreproduced over 75 times; (III) the effect of UGT1A1*28 polymorphism on neutropenia using either dominantmodel (UGT1A1 *1/*1 vs. *1/*28 and *28/*28) or reces-sive model (UGT1A1 *1/*1 and *1/*28 vs. *28/*28) was

Fig. 5 The association between the UGT1A1 *28 or SLCO1B1 c.521T>C polymorphism and dose-normalized SN-38 plasma concentration at the end ofinfusion (90 min) with 9 VCSs of 100 times of VCSs. (a) and (a) show the association between the UGT1A1 *28 polymorphism and dose-normalized SN-38plasma concentration. (c) and (d) show the association between the SLCO1B1 c.521T>C polymorphism and dose-normalized SN-38 plasma concentration. In(a) and (c), each virtual patient was generated from the set of parameters ID 1. In (b) and (d), each virtual patient was generated from the set of parameters ID 5.The top-left panel in each figure shows the results of the target study (20). The others show 9 of the 100 times of VCSs. * p< 0.05, ** p< 0.01, *** p< 0.01.


reproduced over 75 times; (IV) the effect of UGT1A1 *28polymorphism on diarrhea using either dominant modelor recessive model was reproduced over 25 times; (V) theeffect of SLCO1B1 c.521T>C on plasma concentration ofSN-38 was reproduced over 25 times; (VI) the effect ofSLCO1B1 c.521T>C on neutropenia using either domi-nant model (SLCO1B1 521 T/T vs. T/C and C/C) orrecessive model (SLCO1B1 521 T/T and T/C vs. C/C)was reproduced over 25 times; and (VII) the associationbetween biliary index and diarrhea was reproduced over25 times. Under this definition, 19 sets of parameters weresatisfactory in terms of the consistency of more than orequal to 5 of 7 criteria. Especially, ID 2, 10, 11, 12, 14,and 18 were satisfactory with consistency of more than 5criteria.

DISCUSSION

The purpose of this study was to establish a PBPK model thatcould be used to obtain insight into the factors associated withthe side effects caused by irinotecan. First, the key findings werebriefly described. The parameters for the PBPK model wereestimated as the first step in our analyses. Because it was difficultto estimate a large number of unknown parameters, CNMwasnewly employed to optimize unknown parameters. By usingCNM, over 30 sets of parameters, which could reproduce notonly the average blood concentration–time profile of irinotecanand its metabolites, but also urinary and fecal accumulation–time profiles (24), were successfully obtained. The sets of pa-rameters obtained were also evaluated in terms of whether theycould reproduce the results of reported clinical outcomes of

Fig. 5 (continued)


irinotecan-induced side effects such as neutropenia and delayeddiarrhea using a VCS approach. The VCS results indicatedthree important findings: (I) inter-individual variability of pa-tients has the potential to cause a different clinical outcomewitha small number of patients; n ≥ 192 is needed to obtain asignificant result with the effect of SLCO1B1 c.521T>C onthe plasma concentration of SN-38, (II) specific parameterswere strongly correlated with the frequency of reproducingclinical observations, and (III) the biliary index is consideredto be a better biomarker than UGT1A1 *28 polymorphismregarding susceptibility to diarrhea.

PBPK Modeling of Irinotecan and Its Metabolites

First, the contribution of CES1 to the metabolism ofirinotecan to SN-38 in the liver was estimated using CES1AIgG. As shown in Fig. 3, it was found that CES2 plays a keyrole in the metabolism of irinotecan to SN-38 in the liver. Therelative expression level of mRNA for CES2 to CES1 in theintestine is higher than that in the liver (39), suggesting that thecontribution of CES1 to the metabolism of irinotecan is con-sidered small if any in the liver and intestine. Therefore, theCES1 genotype was not considered in the VCS.

Figure 4 shows that multiple sets of parameters in thePBPK model of irinotecan and its metabolites that could re-produce the observed blood concentration–time profiles. ThisPBPK-based optimization was obtained using a CNM (18).Although the same optimization using a Levenberg–Marquardt method, which is a conventional fitting approach,was performed onMATLAB, parameters were not convergedbecause of too many unknown parameters. The computation-al time for CNM optimization with the initial 10,000 sets ofparameters was about 15 min. This result shows that CNM is

a powerful algorithm by which to find multiple sets of param-eters for PBPK models.

Yoshida et al. (19) reported estimated PBPKmodel param-eters for the metabolism of irinotecan and its metabolites thatreproduced observed urinary and fecal accumulation–timeprofiles (24) using CNM. Urinary and fecal accumulationwas also simulated using the 30 sets of parameters obtainedusing CNM (Supplementary Fig. 3). Considering the varianceof the observed accumulation, the simulated accumulationprofiles using these set s of parameters are considered to becomparable with the observed accumulation–time profiles.

VCSs Using Sets of Parameters Obtained by CNMto Examine whether the Clinical Outcome with Regardto the Effect of Genetic Polymorphism on the SN-38Plasma Concentrations and Side Effects Can beProduced

As shown in Fig. 5, even when the virtual populations weregenerated from the same average value for all parameters,different results of VCSwith different virtual populations wereobtained. This result indicates that inter-individual variabilityof patients has the potential to cause a different clinical out-come with such a small number of patients as the 127 adoptedin the target clinical study (20). To calculate the effect of inter-individual variability, 100 VCSs were performed with differ-ent numbers of virtual patients. Figure 6 shows the reproduc-tion frequency of the plasma SN-38 concentration at the endof infusion (90 min) among 100 VCSs with regard to the effectof UGT1A1 *28 (Fig. 6 a) and SLCO1B1 c.521T>C (Fig. 6b), which was estimated by VCSs changing the number ofvirtual patients in each clinical study. In Fig. 6, the reproduc-tion frequency increases according to the increase in the num-ber of virtual population. In this analysis, the reproduction

Fig. 6 The association between the number of virtual patient and the reproduction frequency of the effect of UGT1A1 *28 or SLCO1B1 c.521T>Cpolymorphism on dose-normalized SN-38 plasma concentration at the end of infusion (90 min) among 100 times of VCSs. The reproduction frequency of(a) shows the effect of UGT1A1 *28 polymorphism on dose-normalized SN-38 plasma concentration. The reproduction frequency of (b) shows the effect ofSLCO1B1 c.521T>C polymorphism on dose-normalized SN-38 plasma concentration. The set of parameters ID 1 (×) and 5 (○) were used. The number ofvirtual population in X-axis is shown as a logarithmic scale.


frequency is an index similar to statistical power. The statisti-cal power of 0.8 has been often used as the threshold of itspower.When 0.8 was applied as the threshold of reproductionfrequency, the reproduction frequency was higher than 0.8 atn = 192 for SLCO 1B1 and ID 5, at n = 96 for UGT1A1 andID 1 (Fig. 6). ID 5 used in this analysis could reproduce theeffect of SLCO1B1 polymorphism on the plasma SN-38 con-centration with the highest frequency in all sets of parametersfor n = 127 (Fig. 8). On the other hand, many sets of param-eters could reproduce the effect of UGT1A1 polymorphismwith the highest frequency (Fig. 8). It was thus estimated thatn ≥ 192 is needed to obtain the significant result of the effect ofSLCO1B1 c.521T>C. Generally, it is difficult to perform aclinical study using such large number of patients. On theother hand, large-scale VCSs can overcome the effect of

inter-individual variability and provide a good estimate of trueclinical outcomes.

Some clinical studies have reported that the UGT1A1*28genotype does not affect the frequency of neutropenia (40),even if many other studies have reported this effect. Figure 8shows that the impact of UGT1A1*28 polymorphism on in-creasing the plasma SN-38 concentration and side effects; neu-tropenia was reproduced in more than 80 of the 100 times ofVCS trials with most sets of parameters except for ID 4, 5, 6,15, 24, and 29. The common features of parameters among ID4, 5, 6, 15, 24, and 29 were searched, and these 6 sets werefound to have the lowest values for the ratio (fglu(SN-38)) of met-abolic clearance for glucuronidation of SN-38 to hepatic in-trinsic clearance. Then, the correlation of fglu(SN-38) and thereproduction frequency of the impact of UGT1A1*28

Fig. 7 Comparison between clinical study and virtual clinical study outcomes using 9 different sets of parameters determined by CNM. (a) and (b) show theassociation between dose-normalized SN-38 plasma concentrations at the end of infusion (90min) and the UGT1A1*28 genotype (a) and SLCO1B1 c.521T>Cgenotype (b). The top-left panel in each figure shows the results of the target study (20). The others show 9 sets of parameters of the 30 sets optimized by CNM.* p < 0.05, ** p < 0.01, *** p < 0.01.


polymorphism on increasing the neutropenia was plotted using30 sets of parameters (Supplementary Fig. 4A). SupplementaryFig. 4A shows that each set of parameters in which fglu(SN-38)was more than 0.8 had a high reproduction frequency of morethan 0.8. This is because the effect of UGT1A1 *28 is lesspronounced when the fglu(SN-38) value becomes smaller.Similarly, when the correlation of Rdif,h(SN-38) and the repro-duction frequency of the impact of SLCO1B1 c.521T>C poly-morphism on increasing the neutropenia was plotted using 30sets of parameters (Supplementary Fig. 4B), the set of param-eters having a large Rdif,h(SN-38) could not reproduce the effectof SLCO1B1 c.521T>C on neutropenia with high frequency.Because the contribution of SLCO1B1 to the total hepaticuptake clearance becomes smaller with larger Rdif,h(SN-38),SLCO1B1 c.521T>C polymorphism had little effect on SN-38 plasma exposure.

The target study also found that only heterozygousABCC2 c.-24C>T, but not homozygous carriers reducedthe risk of neutropenia (Fig. 8). As seen in Fig. 8, no set ofparameters could reproduce this result in more than 10 of the100 times of VCSs. We believe that the target study may haveobtained this result by chance, considering only heterozygouscarriers achieved statistical significance. In association withdiarrhea, the target study found that a common SNP inCES1 (rs2244613) caused a lower risk. However, the currentstudy (Fig. 1) suggests that CES2 mainly contributes to themetabolism of irinotecan to SN-38 in the liver and intestine,as discussed previously. The mechanism by which the targetstudy found the effect of CES1 rs2244613 on the occurrenceof diarrhea has not yet been clarified.

Gupta et al. (37) proposed that patients with a high biliaryindex have a risk of diarrhea. By contrast, other clinical studieshave reported that UGT1A1 *28 polymorphism increases therisk of diarrhea (4,5). Figure 8 includes the results of testsaccording to the association between these factors and diar-rhea for 100 times of VCSs among 30 sets of parameters. TheVCSs were performed by assuming the diarrhea is influencedby the unbound AUC of SN-38 in the intestinal epithelia. Asshown in Fig. 8, the sets of parameters for which the frequencyof statistical significance of association of UGT1A1 *28 poly-morphism with diarrhea is high also exhibited a high frequen-cy of significant association between the biliary index anddiarrhea. This result may be explained by the possibility thatUGT1A1 *28 polymorphism affects the increase in the biliaryindex, which has been found clinically (41). However, thefrequency of significant association between the biliary indexand diarrhea was higher than that of the association betweenUGT1A1 *28 polymorphism and diarrhea for all sets of pa-rameters (Fig. 8). Consequently, the biliary index is consideredto be a better biomarker than UGT1A1 *28 for susceptibilityof diarrhea.

Perspective of VCS

In the present study, VCS was performed to evaluate whetherreported clinical studies could be reproduced by the PBPKmodel, which was a Bretrospective^ approach. In the practicalapplication of VCS to new drug development, we have to usea Bprospective^ approach. What kind of procedure should beadopted for this purpose? A PBPK model is first established

Fig. 8 Heat map for the frequency of significance match of the 100 virtual clinical studies to a target clinical study for each of the 30 sets of parameters determinedby CNM. Each cell number shows how many times the statistical significance (p < 0.05) for each association described in the left row among 100 VCSs. Smallfrequency of significance described in the cell is colored in yellow, and the higher frequency of significance is colored in green. The BTarget^ column shows theresults of the target study (20). The BOthers^ column shows the results from other reports than the target study. The references described in BOthers^ areprovided in the Supplementary Text. All the associations except for the association between ABCC2 c.-24C>T genotype and the neutropenia shows somehigher frequency of significance between the genetic polymorphisms and SN-38 plasma concentration, neutropenia, or diarrhea. The association betweenABCC2 c.-24C>T genotype and the neutropenia is very small if any. This figure shows the results of statistical test. The complete sets of results are described inSupplementary Table 5.


for a candidate compound after a Phase I clinical study, usingobserved plasma concentration–time profile data togetherwith many available in vitro experimental data on the binding,metabolism and transport.We then can estimate the exposureof the compound in the target tissues as a function of dose interms of pharmacological and side effects. Next, the relation-ship between the simulated target tissue exposure of the com-pound and the pharmacological effects and side effects aresimulated. To do so, we have to estimate the exposure to causepharmacological effects and side effects using in vivo and in vitrodata. Those may include clinical phase I studies (in the case ofanticancer drugs) and in vitro experiments using human nor-mal cells, cancer cells, gene expressed and knockdown cellsand in vivo animal model experiments. These studies will pro-vide the thresholds of exposure, i.e., the criteria as to whethereffects or side effects occur. Then, various virtual populations(e.g., genetic polymorphism, ethnicity and disease state) aregenerated. Finally, after VCS is performed using a generatedvaried virtual population, the PBPK model is developed andthe relationships of target tissue exposure to the effects andside effects of the compound are estimated. Finally, we will beable to estimate the occurrence probability of effects and sideeffects and the suitable dosage for every virtual populationsuch as population with different genetic polymorphisms, co-administered drugs, different ethnicity and dysfunctions oforgans (liver, kidney).

We generated 1,000,000 virtual people using the ID 2 setof parameters as average values and variabilities (shown asCV) as shown in Supplementary Tables 3 and 4, and simulat-ed unbound AUC values of SN-38 in the plasma andenterocyte. The thresholds of neutropenia and diarrhea wereset for plasma unbound AUC> 26.35 [nM⋅h] and enterocyteunbound AUC > 53.60 [nM⋅h], respectively, the values ofwhich were determined as average thresholds for 100 timesof VCSs using the ID 2 set of parameters. Thus, the probabil-ity of neutropenia and diarrhea were estimated by VCS with1,000,000 virtual people (Supplementary Fig. 5). As a result, itwas estimated that 16.3% and 6.04% of patients would expe-rience neutropenia and diarrhea, respectively. The occur-rence frequency of neutropenia and diarrhea were reportedto be 23.0 ± 3.4% and 10.3 ± 3.0%, respectively, by summa-rizing previous clinical studies (Supplementary Table 6).Comparing the frequency of side effects obtained from VCSto that of reported studies, VCS provided somewhat lowervalues, probably because thresholds values were determinedbased only on the results of the target study.

CONCLUSION

This work successfully obtained multiple sets of PBPKmodel parameters that could reproduce the effects of ge-netic polymorphisms of UGT1A1 *28 and SLCO1B1

c.521T>C on the plasma concentration of SN-38 and sideeffects such as neutropenia and diarrhea using a VCS ap-proach. To optimize the numerous biochemical parame-ters of irinotecan and its metabolites in the PBPK model, aCNM parameter optimization algorithm was introduced.VCS suggested that inter-individual variabilities affect clin-ical outcomes when performed with smaller number ofpatients. The current VCS confirmed the importance ofthe biliary index as a better biomarker of irinotecan-induced diarrhea compared with only UGT1A1 *28 poly-morphism. By contrast, VCS suggested that the impor-tance of polymorphism of other factors (CES1, ABCC2,ABCB1, and ABCG2) should be interpreted carefully.

ACKNOWLEDGMENTS AND DISCLOSURES

The authors sincerely thank Drs. Soo-Jin Kim and KazunobuTsunemoto (RIKEN, Yokohama, Japan) for helping the ex-periments of measuring protein binding and partition toerythrocytes of irinotecan and its metabolites. The authorsalso thank for Dr. Teruko Imai (Kumamoto University,Kumamoto, Japan) for discussions about experiments, Drs.Yoshiko Tomita (Dainippon Sumitomo Pharma Co., Osaka,Japan) and Koji Chiba (Yokohama College of Pharmacy,Yokohama, Japan) for analysis of the activity ratio and inter-individual variability for SLCO1B1 polymorphism, and Drs.Ken-ichi Fujita and Yutaro Kubota (Showa University,Tokyo, Japan) for discussions about the information of clinicalstudies. This work was supported by a Grant-in-Aid forYoung Scientists (B) [16K18972 to K.T.] from the JapanSociety for the Promotion of Science (JSPS) and a Grant-in-Aid for Scientific Research (S) [24229002 to Y.S.] from thesame society. K.T. and Y.S. wrote the article. K.T., K.H., andY.S. designed the research. A.T. performed the experiment.K.T., A.T., and Y.S. analyzed the data. The authors have noconflict of interest to be disclosed.

OpenAccessThis article is distributed under the terms of theCreative Commons Attribution 4.0 International License(http://creativecommons.org/licenses/by/4.0/), which per-mits unrestricted use, distribution, and reproduction in anymedium, provided you give appropriate credit to the originalauthor(s) and the source, provide a link to the CreativeCommons license, and indicate if changes were made.

REFERENCES

1. Marsh S, Hoskins JM. Irinotecan pharmacogenomics.Pharmacogenomics. 2010;11(7):1003–10.

2. Paulik A, Grim J, Filip S. Predictors of irinotecan toxicity andefficacy in treatment of metastatic colorectral cancer. Acta Med(Hradec Kralove). 2012;55(4):153–9.


3. Innocenti F, Kroetz DL, Schuetz E, Dolan ME, Ramírez J, RellingM, Chen P, Das S, Rosner GL, Ratain MJ. Comprehensive phar-macogenetic analysis of irinotecan neutropenia and pharmacoki-netics. J Clin Oncol. 2009;27(16):2604–14.

4. Fujiwara Y, Minami H. An overview of the recent progress inirinotecan pharmacogenetics. Pharmacogenomics. 2010;11(3):391–406.

5. Di Paolo A, Bocci G, Polillo M, Del ReM, Di Desidero T, LastellaM, Danesi R. Pharmacokinetic and pharmacogenetic predictivemarkers of irinotecan activity and toxicity. Curr Drug Metab.2011;12(10):932–43.

6. Martinez-Balibrea E, Abad A, Martínez-Cardús A, Ginés A,Valladares M, Navarro M, Aranda E, Marcuello E, BenavidesM, Massutí B, Carrato A, Layos L, Manzano JL, Moreno V.UGT1A and TYMS genetic variants predict toxicity and responseof colorectal cancer patients treated with first-line irinotecan andfluorouracil combination therapy. Br J Cancer. 2010;103(4):581–9.

7. Ruzzo A, Graziano F, Loupakis F, Santini D, Catalano V, BisonniR, Ficarelli R, Fontana A, Andreoni F, Falcone A, Canestrari E,Tonini G,Mari D, Lippe P, Pizzagalli F, Schiavon G, AlessandroniP, Giustini L, Maltese P, Testa E, Menichetti ET, Magnani M.Pharmacogenetic profiling in patients with advanced colorectalcancer treated with first-line FOLFIRI chemotherapy.Pharmacogenomics J. 2008;8(4):278–88.

8. Han JY, Lim HS, Park YH, Lee SY, Lee JS. Integrated pharma-cogenetic prediction of irinotecan pharmacokinetics and toxicity inpatients with advanced non-small cell lung cancer. Lung Cancer.2009;63(1):115–20.

9. Lara Jr PN, Natale R, Crowley J, LenzHJ, RedmanMW,CarletonJE, Jett J, Langer CJ, Kuebler JP, Dakhil SR, Chansky K, GandaraDR. Phase III trial of irinotecan/cisplatin compared withetoposide/cisplatin in extensive-stage small-cell lung cancer: clinicaland pharmacogenomic results from SWOG S0124. J Clin Oncol.2009;27(15):2530–5.

10. Cha PC, Mushiroda T, Zembutsu H, Harada H, Shinoda N,Kawamoto S, Shimoyama R, Nishidate T, Furuhata T, Sasaki K,Hirata K, Nakamura Y. Single nucleotide polymorphism inABCG2 is associated with irinotecan-induced severemyelosuppression. J Hum Genet. 2009;54(10):572–80.

11. Jada SR, Lim R, Wong CI, Shu X, Lee SC, Zhou Q, Goh BC,Chowbay B. Role of UGT1A1*6, UGT1A1*28 and ABCG2c.421C>A polymorphisms in irinotecan-induced neutropenia inAsian cancer patients. Cancer Sci. 2007;98(9):1461–7.

12. Pillot GA, Read WL, Hennenfent KL, Marsh S, Gao F,Viswanathan A, Cummings K,McLeodHL, Govindan R. A phaseII study of irinotecan and carboplatin in advanced non-small celllung cancer with pharmacogenomic analysis: final report. J ThoracOncol. 2006;1(9):972–8.

13. Sager JE, Yu J, Ragueneau-Majlessi I, Isoherranen N.Physiologically based pharmacokinetic (PBPK) modeling and sim-ulation approaches: a systematic review of published models, ap-plications, and model verification. Drug Metab Dispos.2015;43(11):1823–37.

14. Huang SM, Abernethy DR, Wang Y, Zhao P, Zineh I. The utilityof modeling and simulation in drug development and regulatoryreview. J Pharm Sci. 2013;102(9):2912–23.

15. Yamada A, Maeda K, Kiyotani K, Mushiroda T, Nakamura Y,Sugiyama Y. Kinetic interpretation of the importance ofOATP1B3 and MRP2 in docetaxel-induced hematopoietic toxici-ty. CPT Pharmacometrics Syst Pharmacol. 2014;3:e126.

16. Rose RH, Neuhoff S, Abduljalil K, Chetty M, Rostami-HodjeganA, Jamei M. Application of a physiologically based pharmacokinet-ic model to predict OATP1B1-related variability in pharmacody-namics of Rosuvastatin. CPT Pharmacometrics Syst Pharmacol.2014;3:e124.

17. Fujita K, Masuo Y, Okumura H, Watanabe Y, Suzuki H,Sunakawa Y, Shimada K, Kawara K, Akiyama Y, Kitamura M,Kunishima M, Sasaki Y, Kato Y. Increased plasma concentrationsof unbound SN-38, the active metabolite of irinotecan, in cancerpatients with severe renal failure. Pharm Res. 2016;33(2):269–82.

18. Aoki Y, Hayami K, Sterck HD, Konagaya A. Cluster Newtonmethod for sampling multiple solutions of underdetermined inverseproblems: application to a parameter identification problem inpharmacokinetics. SIAM J Sci Comput. 2014;36:B14–44.

19. Yoshida K, Maeda K, Kusuhara H, Konagaya A. Estimation offeasible solution space using Cluster Newtonmethod: application topharmacokinetic analysis of irinotecan with physiologically-basedpharmacokinetic models. BMC Syst Biol. 2013;7(Supp 3):S3.

20. TeftWA,Welch S, Lenehan J, Parfitt J, Choi YH,Winquist E, KimRB. OATP1B1 and tumour OATP1B3 modulate exposure, toxic-ity, and survival after irinotecan-based chemotherapy. Br J Cancer.2015;112(5):857–65.

21. Hosokawa M, Hirata K, Nakata F, Suga T, Sato T. Species differ-ences in the induction of hepatic microsomal carboxylesterasecaused by dietary exposure to di(2-ethylhexyl) phthalate, a peroxi-some proliferator. Drug Metab Dispos. 1994;22(6):889–94.

22. Imai T, Imoto M, Sakamoto H, Hashimoto M. Identification ofesterases expressed in Caco-2 cells and effects of their hydrolyzingactivity in predicting human intestinal absorption. Drug MetabDispos. 2005;33(8):1185–90.

23. Watanabe T, Kusuhara H, Maeda K, Shitara Y, Sugiyama Y.Physiologically based pharmacokinetic modeling to predicttransporter-mediated clearance and distribution of pravastatin inhumans. J Pharmacol Exp Ther. 2009;328(2):652–62.

24. Slatter JG, Schaaf LJ, Sams JP, Feenstra KL, Johnson MG,Bombardt PA, Cathcart KS, Verburg MT, Pearson LK,Compton LD, Miller LL, Baker DS, Pesheck CV, Lord 3rd RS.Pharmacokinetics, metabolism, and excretion of irinotecan (CPT-11) following I.V. infusion of [(14)C]CPT-11 in cancer patients.Drug Metab Dispos. 2000;28(4):423–33.

25. Cong D, Doherty M, Pang KS. A new physiologically based,segregated-flow model to explain route-dependent intestinal me-tabolism. Drug Metab Dispos. 2000;28(2):224–35.

26. Rodgers T, Leahy D, Rowland M. Physiologically based pharma-cokinetic modeling 1: predicting the tissue distribution of moderate-to-strong bases. J Pharm Sci. 2005;94(6):1259–76.

27. Rodgers T, Rowland M. Physiologically based pharmacokineticmodelling 2: predicting the tissue distribution of acids, very weakbases, neutrals and zwitterions. J Pharm Sci. 2006;95(6):1238–57.

28. Yoshikado T, Yoshida K, Kotani N, Nakada T, Asaumi R,Toshimoto K, Maeda K, Kusuhara H, Sugiyama Y. Quantitativeanalyses of hepatic OATP-mediated interactions between statinsand OATP inhibitors using PBPK modeling with a parameter-optimization method. Clin Pharmacol Ther. 2016;100(5):513–23.

29. Watanabe T, Kusuhara H, Sugiyama Y. Application of physiolog-ically based pharmacokinetic modeling and clearance concept todrugs showing transporter-mediated distribution and clearance inhumans. J Pharmacokinet Pharmacodyn. 2010;37(6):575–90.

30. Fujita K, Sugiura T, Okumura H, Umeda S, Nakamichi N,Watanabe Y, Suzuki H, Sunakawa Y, Shimada K, Kawara K,Sasaki Y, Kato Y. Direct inhibition and down-regulation by uremicplasma components of hepatic uptake transporter for SN-38, anactive metabolite of irinotecan, in humans. Pharm Res.2014;31(1):204–15.

31. DeSesso JM, Jacobson CF. Anatomical and physiological parame-ters affecting gastrointestinal absorption in humans and rats. FoodChem Toxicol. 2001;39(3):209–28.

32. van der Bol JM, LoosWJ, de Jong FA, vanMeerten E, Konings IR,Lam MH, de Bruijn P, Wiemer EA, Verweij J, Mathijssen RH.Effect of omeprazole on the pharmacokinetics and toxicities of


irinotecan in cancer patients: a prospective cross-over drug-druginteraction study. Eur J Cancer. 2011;47(6):831–8.

33. Sharkey I, Boddy AV, Wallace H, Mycroft J, Hollis R, Picton S.Chemotherapy standardisation group of the United KingdomChildren's cancer study group. Body surface area estimation inchildren using weight alone: application in paediatric oncology.Br J Cancer. 2001;85(1):23–8.

34. Myrand SP, Sekiguchi K, Man MZ, Lin X, Tzeng RY, TengCH, Hee B, Garrett M, Kikkawa H, Lin CY, Eddy SM,Dostalik J, Mount J, Azuma J, Fujio Y, Jang IJ, Shin SG,Bleavins MR, Williams JA, Paulauskis JD, Wilner KD.Pharmacokinetics/genotype associations for major cyto-chrome P450 enzymes in native and first- and third-generation Japanese populations: comparison with Korean,Chinese, and Caucasian populations. Clin Pharmacol Ther.2008;84(3):347–61.

35. Murray DM, Burmaster DE. Estimated distributions for total bodysurface area of men and women in the United States. J Expo AnalEnviron Epidemiol. 1992;2(4):451–61.

36. Sai K, Kaniwa N, Itoda M, Saito Y, Hasegawa R, KomamuraK, Ueno K, Kamakura S, Kitakaze M, Shirao K, Minami H,Ohtsu A, Yoshida T, Saijo N, Kitamura Y, Kamatani N,Ozawa S, Sawada J. Haplotype analysis of ABCB1/MDR1blocks in a Japanese population reveals genotype-dependent

renal clearance of irinotecan. Pharmacogenetics. 2003;13(12):741–57.

37. Gupta E, Lestingi TM, Mick R, Ramirez J, Vokes EE, RatainMJ. Metabolic fate of irinotecan in humans: correlation ofglucuronidation with diarrhea. Cancer Res. 1994;54(14):3723–5.

38. Takai S, Matsuda A, Usami Y, Adachi T, Sugiyama T, Katagiri Y,Tatematsu M, Hirano K. Hydrolytic profile for ester- or amide-linkage by carboxylesterases pI 5.3 and 4.5 from human liver. BiolPharm Bull. 1997;20(8):869–73.

39. NishimuraM,Naito S. Tissue-specific mRNA expression profiles ofhuman phase I metabolizing enzymes except for cytochrome P450and phase II metabolizing enzymes. Drug Metab Pharmacokinet.2006;21(5):357–74.

40. Braun MS, Richman SD, Thompson L, Daly CL, MeadeAM, Adlard JW, Allan JM, Parmar MK, Quirke P,Seymour MT. Association of molecular markers with toxicityoutcomes in a randomized trial of chemotherapy for ad-vanced colorectal cancer: the FOCUS trial. J Clin Oncol.2009;27(33):5519–28.

41. Iyer L, Das S, Janisch L, Wen M, Ramírez J, Karrison T, FlemingGF, Vokes EE, Schilsky RL, Ratain MJ. UGT1A1*28 polymor-phism as a determinant of irinotecan disposition and toxicity.Pharmacogenomics J. 2002;2(1):43–7.


Virtual Clinical Studies to Examine the Probability Distribution of … · 2017. 8. 25. · for...

Documents

Transcript of Virtual Clinical Studies to Examine the Probability Distribution of … · 2017. 8. 25. · for...